Locked In Deep Learning. https://teapaw.org/t.co/Lm67j3KO0T
Joined July 2025
- Tweets1.6K
- Following57
- Followers390
- Likes6.1K
Pinned Tweet
Introducing,
PulmoScan AI - A high-sensitivity Pneumonia Screening tool powered by Deep Learning
> With GradCAM visuals.
> DenseNet121 - misses rarely 1 cases with less false alarms
> Higher recall ~99.7% Vs. ResNet50 ~97%
App:pulmoscan.streamlit.app
Repo Link in comment.
yo chat checkout this RAG based Document Intelligence Chatbot - pretty simple
#BuildingInPublic #BuildingWithLearning #RAG #AI
# Built a Document Intelligence Chatbot!
> RAG pipeline: FastAPI + Gemini Embeddings + Qdrant Cloud + Groq LLM
> Upload PDFs/DOCX/TXT, ask questions, get answers with cited sources — all free tier, no PyTorch needed!
> Cloudflare Workers frontend • Deployed on Render
1
1
101
yo chat Updates on my RAG based Document Intelligence chatbot
> Here is the live link for this project : document-chatbot.omm512462.w…
> Github Repo : github.com/Om-codex/document…
2
25
Updates on RAG-based Document Chatbot:
- finally after wasting 2 days due to festival
- the frontend is ready and working fine
- the Groq llm responds to the chunks and the given question
- also returns the chunks for debugging
- soon final project will be deployed
#BuildInPublic
5
9
114
Updates on RAG based Document Chatbot
- Model Caching to avoid repeated loading models
- Centralized Configuration
- better logging, input validation
- error recovery
- and the final frontend UI will be completed by today
Happy Diwali Everyone 🤍🌟
#AI #BuildInPublic #RAG #LLMs
1
10
91
Updates on RAG based Document Chatbot
- the service layer is working
- refactored each routes on service layer
Next milestones:
- model caching
- better loggin
- error recovery
- integration tests
Then finally the AI-agent will build frontend for me :)
3
47
building this project feels so good like,
- whole RAG engine is now working
- all FastAPI 5 endpoints are working correctly
- the only fix: handle empty document IDs in chat requests
now fastapi also returns answer + retrieved chunks(given to llm) - also helps in debugging
2
3
40
Fastapi integration is started on this project btw it's a just small learning project
been exploring rag and building a document chatbot.
document ingestion & chunking
embeddings + chromadb
semantic retrieval
qa chain with langchain
rag engine is working. now moving to fastapi + frontend integration.
building while learning.
2
24
it's cool to build chatbots using langchain :)
• building many things on langchain-models using langchain
• static & dynamic prompting
• structured outputs using TypedDict,Pydantic & json schema
• about output parsers
2
220
Just implemented Self-Attention from scratch after building Seq2Seq + Attention
Started from encoder-decoder, struggled with EOS bugs, fixed training logic, added attention… and finally saw how models actually “focus” on words
Now moving towards Transformers & GPT
Grind It 🔥
2
195
Om. retweeted
A recent project that I have built and am currently working on : Vigil , a real-time log ingestion and threat detection Platform. The system ingests logs , aggregates them into time-windowed feature sets and runs a two- stage pipeline. Explanation and architecture diagram below
The actual process to calculate the contextual embeddings for self attention in Transformers
8
1,148
Chapter 3 : Mind your own business
> There's a difference between a profession & your own business.
> Rich people focus on their assets not on their income statements.
>First grow your assets column, build business & then only buy luxuries.
6
83
